I thought, without trying it, that the hit from the second match would be compensated for by not using $&, but benchmark tests I tried indicate that that is not the case. I didn't see any additional hit as a result of using $& at all!

I guess the moral is: try the simple stuff, but validate assumptions.